Update app.py

app.py

@@ -1,6 +1,6 @@
 import os
-import spaces
 import re
+import spaces
 import shutil
 import zipfile
 import torch
@@ -41,8 +41,6 @@ def correct_typography(text):
     corrected_text = re.sub(r"([?!:;])(?=\w)", r"\1 ", corrected_text)  # Ajout d'un espace après !, ?, : et ; si nécessaire
     corrected_text = re.sub(r"(?<!\d) (\.)", r"\1", corrected_text)  # Suppression de l'espace avant un point
     corrected_text = re.sub(r"(\.) (?=\w)", r". \2", corrected_text)  # Ajout d'un espace après un point si nécessaire
-    
-    # Appliquer la correction uniquement si le texte a changé
     return corrected_text.strip() if corrected_text != text else text
 
 # -------------------------------------------------
@@ -50,11 +48,10 @@ def correct_typography(text):
 # -------------------------------------------------
 @spaces.GPU(duration=120)
 def transcribe_audio(audio_path):
-    import os
     if not audio_path:
         return "Aucun fichier audio fourni", None, [], "", ""
     
-    file_name = os.path.basename(audio_path).rsplit('.', 1)[0]  # Extraire le nom sans extension
+    file_name = os.path.basename(audio_path).rsplit('.', 1)[0]  # Extraction du nom sans extension
     
     result = pipe(audio_path, return_timestamps="word")
     words = result.get("chunks", [])
@@ -67,39 +64,76 @@ def transcribe_audio(audio_path):
     transcription_with_timestamps = " ".join([f"{w[0]}[{w[1]:.2f}-{w[2]:.2f}]" for w in word_timestamps])
     
     return raw_transcription, word_timestamps, transcription_with_timestamps, audio_path, file_name
-    if not audio_path:
-        return "Aucun fichier audio fourni", None, [], ""
-    
-    result = pipe(audio_path, return_timestamps="word")
-    words = result.get("chunks", [])
-    
-    if not words:
-        return "Erreur : Aucun timestamp détecté.", None, [], ""
-    
-    raw_transcription = " ".join([w["text"] for w in words])
-    word_timestamps = [(w["text"], w["timestamp"][0], w["timestamp"][1]) for w in words]
-    transcription_with_timestamps = " ".join([f"{w[0]}[{w[1]:.2f}-{w[2]:.2f}]" for w in word_timestamps])
-    
-    return raw_transcription, word_timestamps, transcription_with_timestamps, audio_path
 
 # -------------------------------------------------
 # 3. Enregistrement des segments définis par l'utilisateur
 # -------------------------------------------------
-def save_segments(table_data):
+def save_segments(table_data, zip_name):
+    print("[LOG] Enregistrement des segments définis par l'utilisateur...")
+    formatted_data = []
+    confirmation_message = "**📌 Segments enregistrés :**
+"
+
+    for i, row in table_data.iterrows():
+        text, start_time, end_time = row["Texte"], row["Début (s)"], row["Fin (s)"]
+        segment_id = f"{zip_name}_seg_{i+1:02d}"
+
+        try:
+            start_time = str(start_time).replace(",", ".")
+            end_time = str(end_time).replace(",", ".")
+
+            if not start_time.replace(".", "").isdigit() or not end_time.replace(".", "").isdigit():
+                raise ValueError("Valeurs de timestamps invalides")
+
+            start_time = float(start_time)
+            end_time = float(end_time)
+
+            if start_time < 0 or end_time <= start_time:
+                raise ValueError("Valeurs incohérentes")
+
+            formatted_data.append([text, start_time, end_time, segment_id])
+            log_message = f"- `{segment_id}` | **Texte** : {text} | ⏱ **{start_time:.2f}s - {end_time:.2f}s**"
+            confirmation_message += log_message + "
+"
+            print(f"[LOG] {log_message}")
+
+        except ValueError as e:
+            print(f"[LOG ERROR] Erreur de conversion des timestamps : {e}")
+            return pd.DataFrame(), "❌ **Erreur** : Vérifiez que les valeurs sont bien des nombres valides."
+
+    return pd.DataFrame(formatted_data, columns=["Texte", "Début (s)", "Fin (s)", "ID"]), confirmation_message, zip_name
     formatted_data = []
     for i, row in table_data.iterrows():
         formatted_data.append([row["Texte"], float(row["Début (s)"]), float(row["Fin (s)"])] )
-    return pd.DataFrame(formatted_data, columns=["Texte", "Début (s)", "Fin (s)"])
+    return pd.DataFrame(formatted_data, columns=["Texte", "Début (s)", "Fin (s)"]), zip_name
 
 # -------------------------------------------------
 # 4. Génération du fichier ZIP avec correction typographique
 # -------------------------------------------------
 def generate_zip(metadata_state, audio_path, zip_name):
+    if isinstance(metadata_state, tuple): 
+        metadata_state = metadata_state[0]
+    
     if metadata_state is None or metadata_state.empty:
+        print("[LOG ERROR] Aucun segment valide trouvé pour la génération du ZIP.")
         return None
     
-    if not zip_name.strip():
-        zip_name = "processed_audio"
+    zip_folder_name = f"{zip_name}_dataset"
+    zip_path = os.path.join(TEMP_DIR, f"{zip_folder_name}.zip")
+    metadata_csv_path = os.path.join(TEMP_DIR, "metadata.csv")
+    
+    metadata_state["ID"] = [f"{zip_name}_seg_{i+1:02d}" for i in range(len(metadata_state))]
+    metadata_state["Texte"] = metadata_state["Texte"].apply(correct_typography)
+    metadata_state["Commentaires"] = ""
+    metadata_state.to_csv(metadata_csv_path, sep="|", index=False)
+    
+    with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
+        zf.write(metadata_csv_path, "metadata.csv")
+    
+    print("[LOG] Fichier ZIP généré avec succès.")
+    return zip_path
+    if metadata_state is None or metadata_state.empty:
+        return None
     
     zip_folder_name = f"{zip_name}_dataset"
     zip_path = os.path.join(TEMP_DIR, f"{zip_folder_name}.zip")
@@ -112,15 +146,6 @@ def generate_zip(metadata_state, audio_path, zip_name):
     
     with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
         zf.write(metadata_csv_path, "metadata.csv")
-        original_audio = AudioSegment.from_file(audio_path)
-        
-        for i, row in metadata_state.iterrows():
-            start_ms, end_ms = int(row["Début (s)"] * 1000), int(row["Fin (s)"] * 1000)
-            segment_audio = original_audio[start_ms:end_ms]
-            segment_filename = f"{zip_name}_seg_{i+1:02d}.wav"
-            segment_path = os.path.join(TEMP_DIR, segment_filename)
-            segment_audio.export(segment_path, format="wav")
-            zf.write(segment_path, segment_filename)
     
     return zip_path
 
@@ -131,17 +156,16 @@ with gr.Blocks() as demo:
     gr.Markdown("# Application de Découpe Audio")
     metadata_state = gr.State(init_metadata_state())
     audio_input = gr.Audio(type="filepath", label="Fichier audio")
-    zip_name = gr.Textbox(label="Nom du fichier ZIP", placeholder="Nom personnalisé", interactive=True)
+    zip_name = gr.Textbox(label="Nom du fichier ZIP", interactive=True)
     raw_transcription = gr.Textbox(label="Transcription", interactive=True)
     transcription_timestamps = gr.Textbox(label="Transcription avec Timestamps", interactive=True)
-    table = gr.Dataframe(headers=["Texte", "Début (s)", "Fin (s)"], datatype=["str", "str", "str"], row_count=(5, "dynamic"))
+    table = gr.Dataframe(headers=["Texte", "Début (s)", "Fin (s)"], datatype=["str", "str", "str"], row_count=(1, "dynamic"))
     save_button = gr.Button("Enregistrer les segments")
     generate_button = gr.Button("Générer ZIP")
     zip_file = gr.File(label="Télécharger le ZIP")
-    word_timestamps = gr.State()
     
-    audio_input.change(transcribe_audio, inputs=audio_input, outputs=[raw_transcription, word_timestamps, transcription_timestamps, audio_input, zip_name])
-    save_button.click(save_segments, inputs=table, outputs=[metadata_state, zip_name])
+    audio_input.change(transcribe_audio, inputs=audio_input, outputs=[raw_transcription, transcription_timestamps, audio_input, zip_name])
+    save_button.click(save_segments, inputs=[table, zip_name], outputs=[metadata_state, zip_name])
     generate_button.click(generate_zip, inputs=[metadata_state, audio_input, zip_name], outputs=zip_file)
 
 demo.queue().launch()